01What Claude Code parallel agents are and how they work

Claude Code parallel agents let one person run several AI agents side by side in the desktop app. Each agent works on its own job in its own session, while you watch everything in one window.

Before this update, if you wanted three agents, you had to open three windows or terminals. You had to remember which window was doing what, and one agent might finish and wait for you without you noticing. The new way shows all sessions in a single sidebar.

Sessions are like one worker at one desk. Now you can have many workers at many desks. Each agent reports back to your main session when it's done.

You see the result, not all the messy work in between.

02Three main features that make parallel agents different

Many tools can run multiple agents. But this update has three things that make it stand out.

  • Nested sessions: When you ask for several tasks at once, Claude starts a separate session for each. Each helper reports its result back to your main session. Your main thread stays clean and easy to read.
  • Follow-up work and /btw: The sidebar shows what's completed, what's migrated, and what needs your input. You can also type /btw (forward slash by the way) to ask a side question. That question opens in its own tab, and the main session keeps going.
  • Dashboard: A sidebar shows which sessions are working and which need you. Inside each session, there's a live to-do list. When Claude needs you, a recap card sums up what just happened so you don't scroll back.

In short, you stop babysitting your agents. You start checking in on them like a barista watching a row of orders.

03How to set up parallel sessions in Claude Code

Getting started with parallel agents is simple. Here's what you need to know.

  1. Open the new session screen. It keeps your recent folders one click away, so you can pick up where you left off fast.
  2. Save a setup as an environment. An environment is a safe spot for Claude to work from. It could be a few folders together, another computer you connect to (an SSH host), or the cloud.
  3. Turn on a work tree for each folder. A work tree gives each helper its own copy of the folder. That way, agents can't trip over each other's changes.
  4. Start your main session. From there, you can launch parallel sessions for each task.

That's the whole setup. You can choose your environment, switch between local, cloud, or a new environment, and see what's blocked and where.

04How to use parallel agents with the split, watch, collect method

Julian Goldie shows a simple way to use many agents without getting overwhelmed. He calls it the many hands engine. It runs in three moves.

Split: Take one big job and cut it into pieces that don't need each other. For example, writing a Substack post and finding images for it can run side by side. If piece two needs piece one to finish first, they can't be split.

Watch: Don't read every line. Just watch the sidebar. See who's working and who needs you.

Read the recap card when it appears. Your job is checking, not doing.

Collect: Let all results come back to one place. Your main session is home base. All helpers report there, so you end with one clean thread with every answer.

Julian uses this for Goldie agency clients. He asks one Claude to research five topics. Instead of one giant wall of text, he gets five reports dropped back into the main thread.

05Results and comparisons from the episode

Anthropic didn't publish speed tests with this update. They didn't say it's three times faster. So you have to train your head around how it works to see benefits.

The old way had two problems. First, too many windows you can't see in one place. Second, if you asked Claude to do many little jobs, it crammed them into one long thread.

That thread got huge, hard to read, and hard to find answers in.

The new way solves that. Nested sessions keep the main thread clean. The sidebar shows everything at a glance.

Work trees prevent agents from breaking each other's stuff.

Julian uses Claude Code for AI profit boarding guides, Substack posts, and SEO briefs. None of that is code. He says anyone can use it now because the desktop app makes it friendly.

06Tips, limits, and mistakes to avoid with parallel agents

Here are some things to keep in mind.

  • Don't split dependent tasks: If one piece needs another to finish first, they aren't ready to split. Only split pieces that can run on their own.
  • Don't read every line: Your job is to watch the sidebar and check in, not to read all the work. Use the recap card.
  • Use work trees to avoid conflicts: If one session updates sales follow-up emails and another rewrites scripts, they stay in their own lanes with their own copies.
  • More sessions don't mean more mess: The old way was messier with more windows and tabs. The new way organizes everything in one sidebar.
  • Speed gains aren't guaranteed: Anthropic didn't publish benchmarks. You have to learn the workflow to see time savings.

If you need help setting this up for your business, Julian mentions resources inside his AirProfit Boardroom community.

07Who Claude Code parallel agents are for and why they matter

This update is for anyone who uses AI agents and finds waiting to be the bottleneck. The agent is smart and does good work, but you can only watch one at a time. You ask, you wait, you check, you ask again.

You're the slow part.

Now you can run a whole row of agents side by side. One writes a guide, another does research, a third cleans up notes. You sit in one window and see who's busy and who's waiting on you.

You don't need to be a coder. Claude Code works on folders, and a folder can hold documents, notes, spreadsheets, or scripts. Anyone can use it with the desktop app.

This matters because it turns Claude Code into a system that works like a small team of AI agents without you losing track. You split work, watch progress, and collect results in one place.